feat(ci): support manylinux build for cpu / cuda (#899)

.github/workflows/ci.yml

@@ -66,14 +66,23 @@ jobs:
     if: github.event_name != 'pull_request'
     needs: tests
     runs-on: ${{ matrix.os }}
+    container: ${{ matrix.container }}
     strategy:
       matrix:
-        os: [macos-latest, ubuntu-latest]
+        binary: [aarch64-apple-darwin, x86_64-manylinux2014, x86_64-manylinux2014-cuda117]
         include:
           - os: macos-latest
             target: aarch64-apple-darwin
+            binary: aarch64-apple-darwin
           - os: ubuntu-latest
             target: x86_64-unknown-linux-gnu
+            binary: x86_64-manylinux2014
+            container: quay.io/pypa/manylinux2014_x86_64
+          - os: ubuntu-latest
+            target: x86_64-unknown-linux-gnu
+            binary: x86_64-manylinux2014-cuda117
+            container: sameli/manylinux2014_x86_64_cuda_11.7
+            build_args: --features cuda
 
     env:
       SCCACHE_GHA_ENABLED: true
@@ -114,17 +123,17 @@ jobs:
       - run: bash ./ci/prepare_build_environment.sh
 
       - name: Bulid release binary
-        run: cargo build --release --target ${{ matrix.target }} --package tabby
+        run: cargo build ${{ matrix.build_args }} --release --target ${{ matrix.target }} --package tabby
 
       - name: Rename release binary
-        run: mv target/${{ matrix.target }}/release/tabby tabby_${{ matrix.target }}
+        run: mv target/${{ matrix.target }}/release/tabby tabby_${{ matrix.binary }}
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v3
         with:
           retention-days: 3
-          name: tabby_${{ matrix.target }}
-          path: tabby_${{ matrix.target }}
+          name: tabby_${{ matrix.binary }}
+          path: tabby_${{ matrix.binary }}
 
   pre-release:
     if: github.event_name == 'push'

ci/prepare_build_environment.sh

@@ -4,6 +4,23 @@ if [[ "$OSTYPE" == "darwin"* ]]; then
   brew install protobuf
 fi
 
+install_protobuf_centos() {
+  PB_REL="https://github.com/protocolbuffers/protobuf/releases"
+  curl -LO $PB_REL/download/v3.15.8/protoc-3.15.8-linux-x86_64.zip
+  unzip protoc-3.15.8-linux-x86_64.zip -d /usr
+  rm protoc-3.15.8-linux-x86_64.zip
+}
+
 if [[ "$OSTYPE" == "linux"* ]]; then
-  sudo apt-get -y install protobuf-compiler libopenblas-dev
+  if command -v apt-get ; then
+    sudo apt-get -y install protobuf-compiler libopenblas-dev
+  else
+    # Build from manylinux2014 container
+    yum -y install openblas-devel perl-IPC-Cmd unzip curl
+
+    # Disable safe directory in docker
+    git config --system --add safe.directory "*"
+
+    install_protobuf_centos
+  fi
 fi

crates/llama-cpp-bindings/build.rs

@@ -25,6 +25,7 @@ fn main() {
     }
     if cfg!(feature = "cuda") {
         config.define("LLAMA_CUBLAS", "ON");
+        config.define("CMAKE_POSITION_INDEPENDENT_CODE", "ON");
         println!("cargo:rustc-link-search=native=/usr/local/cuda/lib64");
         println!("cargo:rustc-link-lib=cudart");
         println!("cargo:rustc-link-lib=culibos");